.png){kind=icon}
.png){kind=icon}
Return
A Graph-Based Approach for Missing Sensor Data Imputation
.png){kind=icon}
PRE
AI
DOI：10.1109/JSEN.2021.3106656.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The Internet of Things (IoT) oriented intelligent services require high-quality sensor data delivery in the wireless sensor networks (WSNs). However, either due to the sensor malfunctions and commutation errors or simply due to the expensive overhead for making full data forwarding, data corruption and loss is relatively common in WSNs, which adversely affects the data quality and the further decisions taking from data. Motivated by the emerging field of graph signal processing (GSP), we propose to impute the missing values in wireless sensor networks based on the topological information carried in the product graph. The proposed solution captures the joint time-space dependencies among the sensor data through a spatiotemporal (ST) graph, which is a time-vertex graph constructed by taking a strong product of a temporal graph and a spatial sensor network graph. Then, the sensory data are mapped onto the vertices of the ST graph and the spatial-temporal nature of sensor data can be further characterized by the notion of smoothness used in GSP. Moreover, instead of imputing with a given spatial graph, we propose a graph learning-based imputation framework to infer underlying space dependencies between the sensors and thus enhance the data imputation performances. Finally, we validate the proposed recovery method using real-world sensor network datasets. The results demonstrate the superior performance of our proposed graph-based method in sensor data imputation, especially when massive sensor data are lost.
Keywords:
Sensors
Spatiotemporal phenomena
Wireless sensor networks
Sensor phenomena and characterization
Intelligent sensors
Internet of Things
Data models
Graph signal processing
missing data recovery
product graph
wireless sensor networks
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.